In our last reading group, dida’s Machine Learning Scientists discussed the paper (arxiv.org/pdf/2501.12948) from deepseek about the DeepSeek-R1 model and their approach of using reinforcement learning directly without prior fine-tuning.
OpenAI’s new o1 model boosts accuracy with chain-of-thought reasoning, excelling in complex tasks like IMO challenges (83% vs. 13% for GPT-4o) and competitive programming.
Here you can read our summary for the o1 model: shorturl.at/NzjfD
#AI #OpenAI #ML #innovation
NVIDIA Shrinks LLama3.1 8B to 4B with Pruning and Distillation
NVIDIA's latest research reduces LLama3.1 8B to 4B parameters by pruning 50% of its layers. Retrained with 40x fewer tokens, the model sees a 16% MMLU score boost. Read more: arxiv.org/pdf/2407.14679
#AI #research
We would like to present a paper on Physics-Informed Neural Networks (PINNs) that use physical laws to improve neural network training. It features a recent PyTorch implementation demonstrating PINNs by calculating gravity from thrown ball data.
sciencedirect.com/science/ar…
#ML
We recently examined MambaVision, a new backbone combining Vision Transformers (ViT) with Mamba. This integration boosts scores and performance for tasks like object detection and instance segmentation.
More details here: arxiv.org/abs/2407.08083
#machinelearning #tech #ML #AI
Visual Instruction Tuning
Learn about "Visual Instruction Tuning": GPT-4 generates multimodal instruction tracking data. ViT-L/14 integration with Vicuna enables image comprehension and conversational applications.
Read more: arxiv.org/abs/2304.08485
dida will be at #ICLR2024 with two papers in Vienna this year. Check them out in case you are there.
On my way to #ICLR2024. I'm looking forward to interesting discussions. Let me know if you're up for a coffee!
Wed: Fast and unified path gradient estimators for normalizing flows (t.ly/kRlaT)
Thu: Improved sampling via learned diffusions (t.ly/V131m)
𝐒𝐮𝐩𝐞𝐫𝐆𝐥𝐮𝐞: In our last reading group we discussed a paper called "#SuperGlue: Learning Feature Matching with Graph Neural Networks"
𝐆𝐢𝐭𝐇𝐮𝐛 𝐒𝐨𝐮𝐫𝐜𝐞 𝐂𝐨𝐝𝐞: github.com/magicleap/SuperGl…
𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐏𝐚𝐩𝐞𝐫: arxiv.org/pdf/1911.11763.pdf
Follow us on X for more interesting #ML papers that you can practically apply to your projects.
xitter.cf/dida_ML
Celebrating 5 years with the dida-conference!🎂
We had an amazing day filled with machine learning (#ML) talks, panels, and workshops featuring top organizations.
We captured some moments of it and turned them into a short video.
𝐯𝐢𝐝𝐞𝐨: youtube.com/watch?v=y3Le5TGp…
It's really interesting to understand more about the inner workings of LLMs in this way.
We also discussed the more general question of how difficult it is to define single independent "facts" and how changing single facts quickly leads to logical contradictions.
3/
Link to the paper: rome.baulab.info/
For more summaries of interesting research papers, make sure to follow us here on 𝕏 so you don't miss out on any updates.
#AI #ML
